OpenAI’s ChatGPT-4.5 has successfully passed a Turing Test-style evaluation, convincing 73% of participants that it was human. Conducted by UC San Diego, the study found ChatGPT-4.5 outperformed earlier models like GPT-4.0 and ELIZA by effectively mimicking subtle language cues. While the model answered well to questions about human experiences, it struggled with current events and real-time information. The evaluation involved various prompts, including one designed to capture the vernacular of a young, introverted internet user. The study highlighted the social and economic implications of advanced AI, noting potential risks like misinformation and social engineering. As AI technology evolves, understanding its human-like capabilities becomes increasingly relevant. The results of this study indicate significant progress in creating machines that can engage in lifelike conversations, raising questions about the implications of AI that can convincingly mimic human interaction.

Source 🔗